Thesis: Adaptive LiDAR Sampling and Depth Completion using Ensemble Variance
My research interest are in the field of Computer Vision (CV) and Deep Learning (DL). Especially, I'm working on improving depth acquisition using adaptive & image-driven sampling and reconstruction algorithms.
We propose a generic ensemble-based method that may use any depth-completion learning algorithm, as the basis for an adaptive LiDAR sampling scheme. This method is based on choosing depth samples proportionally to the variance between different given predictors. In addition, we use the given algorithms as a "black-box", improving their overall performance. Our method allows a reduction by a factor of 4–10 in the number of measurements required to attain the same accuracy.
​
​
​
​
​
​
​
​
​
​
​
​
​
​
In addition, I participated in another work on adaptive depth completion, that was accepted to ICRA2020: "Super-Pixel Sampler: a Data-driven Approach for Depth Sampling and Reconstruction".
https://www.vision-and-sensing.com/post/icra2020-conference
​
As a side project, I'm a part of a group dealing with drones detection in an urban area, using neural networks.
